Automatic Human Age Estimation System for Face Images

INTRODUCTION: With the development of smart devices, such as smart phones and smart televisions, natural user interfaces (NUIs) become increasingly attractive. In addition, with the vigorous research on three-dimensional (3D) video processing techniques on 3DTV, 3DTV NUIs can be also considered. NUIs offer the advantage of natural interaction with a system using predefined actions and/or physical human characteristics. Along with gesture and speech, the human face, which contains substantial information about a person, has also been widely used for human-machine interaction in many applications including face-based human identification, gender classification, age estimation, facial expression recognition, and race classification. Among these, age estimation using facial images is becoming increasingly important. In addition, age estimation is used for face recognition invariant to age progression. Because of the changing of facial characteristics, such as facial shape and skin detail, according to age progression, face recognition performance is less effective if it neglects age effects. In this case, age estimation can serve as a complement to the primary biometric feature of face. The age estimation and synthesis have been also used for finding lost children. Using age estimation and synthesis, the updated face appearance of lost children after several years can be predicted. Age estimation can be used to prevent children from accessing adult websites and restricted videos and from buying tobacco from automated vending machines. Based on human perception, we can see that there are some differences between men and women in terms of producing facial age features. For example, an adult man can have a beard and rough skin surface, whereas a woman does not have a beard and tends to have smoother skin compared to a man. This suggests that gender can have effects on age estimation. In previous studies, gender is recognized by voice, 3D body shape, or face image. In different ways, they show that gender recognition accuracy is affected by age regions (where the frequency components are extracted), which are determined by the facial feature points.